Rikorose
Zero-lookahead model
Greetings! Thank you again for this amazing work!
Currently the models in models folder either have some kind of lookahead or are a little outdated, as you mentioned in the #391
In the LADSPA plugin README.md you mention that it uses new zero-lookahead model. Could you share this model as a separate onnx.tar.gz file? I have tried searching through the source code for the plugin to somehow get it, but ultimately failed. If it is possible, could you point me in the right way?
Edit: It seems from the source that you added the possibility to use either regular DeepFilterNet3 model or the DeepFilterNet3_ll for the LADSPA plugin. How did you made possible the low latency setting with just the regular DeepFilterNet3?
On the similar note, I am also planning to try and train your model on my own data and also enforce the zero-lookahead in the training. Will simply changing the train config to have df_lookahead and conv_lookahead set to 0 be enough? Or should I do something else?
P.S. Due to being on MacOS I am not currently able to use the plugins. I noticed that you disabled MacOS builds for the plugin, did they not work?
